The editor Michael Carl first authors a thoughtful prelude to the four main parts of the volume. The chapter examines the methodological development in the empirical studies of TPR and introduces the main structure of the book. In the past, most TPR researchers turned to the Think Aloud Protocol (TAP) for a rigorous analysis of the translation process. Recently, many technologies and data analysis tools have been available to the research community, such as eye-tracking, key-logging, electroencephalography (EEG) and functional Magnetic Resonance Imaging (fMRI). The recent development is mainly characterised by three trends. First, the availability of “sensor technologies” (p. xviii) facilitates the collection of enormous amounts of data. For example, the Translation Process Research Database (TPR DB) started as a limited data pool and is now “the largest set of publicly available translation process research data” (p. xviii). Secondly, the latest technologies make higher sampling rates possible, which calls for new interpretation models. Thirdly, TPR also touches upon MT quality control and evaluation. Later in this chapter, the editor outlines the four parts of the volume: the measurements for post-editing activities, word translation entropy (HTra) which refers to “the number of alternative translations for a single source word” (Schaeffer et al. 2016), translation segmentation and difficulties, and a post-cognitivist perspective of TPR. The first section of the volume relates to translation technology, quality and effort. In this section, Do Carmo (p. 3-38) first proposes editing as the link between TPR and MT and describes a set of routines to study editing, such as Translation Edit (Error) Rate (TER). However, these metrics are designed to evaluate what goes wrong in the MT output, and sometimes different edits may be made for the same error (Snover et al. 2006). The edit distances should be assessed based on the actual editing actions instead of MT errors. The approach focusing on edits made by translators may be called Human Edit Rate. Do Carmo uses a word processor and two keyloggers to see how accurately these tools can record the revisions and how these editing actions reveal various editing patterns. Finally, Do Carmo calls for the interaction between MT and TPR to better understand the complexity of translation. Huang and Carl (p. 39-56) introduce a new measure called Word-based Human Edit Rate (WHER) to assess post-editing effort. Unlike previous metrics, WHER can predict the post-editing effort per word. In the study, there is a match between an editing action, the corresponding word in the target text (TT) and the source text (ST) counterpart. Furthermore, eye-tracking and keystroke-logging technologies are used to document how the 21 student translators post-edit audiovisual texts. The findings suggest WHER correlates with keystroke activities, gaze data and lexical variation. This correlation indicates that WHER is a reliable index of post-editing effort. Cumbreño and Aranberri (p. 57-80) explore how simple metrics regarding post-editing effort differ and whether the metrics can reveal the effort variation of translators when they identify and correct various errors. In the study, professional translators are recruited and are requested to post-edit a group of sentences where each contains one error. The simple metrics adopted in the research include the measurements of temporal effort, which include total time and editing time, measures of cognitive effort, which involve pause time, editing pause time, initial pause, final pause, pause count, editing pause count and perceived effort as well as quantification of technical effort that includes keystrokes and Human-targeted Translation Edit (Error) Rate (HTER). After comparing different metrics, they obtain empirical evidence that none correlate except total time and pause count. In addition, the metrics don’t vary according to …
Parties annexes
Bibliography
- Krings, Hans P. (2001): Repairing Texts: Empirical Investigations of Machine Translation Post-Editing Processes. Kent: Kent State University Press.
- Schaeffer, Moritz and Carl, Michael (2014): Measuring the Cognitive Effort of Literal Translation Processes. In: Ulrich Germann, Michael Carl, Philipp Koehn, Germán Sanchis-Trilles, Francisco Casacuberta, Robin Hill and Sharon O’Brien, eds. Proceedings of the Workshop on Humans and Computer-Assisted Translation (HaCaT). (14th Conference of the European Chapter of the Association for Computational Linguistics, Gothenburg, 26 April 2014). Stroudsburg: ACL, 29-37. https://research.cbs.dk/en/publications/proceedings-of-the-workshop-on-humans-and-computer-assisted-trans.
- Schaeffer, Moritz, Dragsted, Barbara, Hvelplund, Kristian Tangsgaard, Balling, Laura Winther and Carl, Michael (2016): Word Translation Entropy: Evidence of Early Target Language Activation During Reading for Translation. In: Michael Carl, Srinivas Bangalore and Moritz Schaeffer, eds. New Directions in Empirical Translation Process Research: Exploring the CRITT TPR-DB. Cham: Springer, 183-210.
- Snover, Matthew, Dorr, Bonnie, Schwartz, Richard, Micciulla, Linnea and Makhoul, John. (2006): A Study of Translation Edit Rate with Targeted Human Annotation. In: Proceedings of the 7th Conference of the Association for Machine Translation of the Americas: Technical Papers. (7th Conference of the Association for Machine Translation in the Americas, Cambridge, 8-12 August 2006). Cambridge: AMTA, 223-231. https://aclanthology.org/2006.amta-papers.25.pdf.
- Sun, Sanjun (2015): Measuring Translation Difficulty: Theoretical and Methodological Considerations. Across. 16(1): 29-54. https://doi.org/10.1556/084.2015.16.1.2.
- Sun, Sanjun and Shreve, Gregory M. (2014): Measuring Translation Difficulty: An Empirical Study. Target. 26(1): 98-127. https://doi.org/10.1075/target.26.1.04sun.